Finite Elements in Analysis and Design 47 (2011) 621–634

Contents lists available at ScienceDirect

Finite Elements in Analysis and Design

journal homepage: www.elsevier.com/locate/finel

Finite element model calibration effects on the earthquake response

of masonry arch bridges
Barıs- Sevim a,b,n, Alemdar Bayraktar b, Ahmet Can Altunis-ik b, Sezer Atamtürktür c, Fatma Birinci b
a
Gümüs- hane University, Department of Civil Engineering, 29000 Gümüs- hane, Turkey
b
Karadeniz Technical University, Department of Civil Engineering, 61080 Trabzon, Turkey
c
Clemson University, Department of Civil Engineering, Clemson, SC 29634, USA

a r t i c l e in f o abstract

Article history: The focus of this paper is to illustrate the importance of model calibration and in situ vibration testing by
Received 18 November 2009 comparing the finite element model predictions of the earthquake response of the two historical arch
Received in revised form bridges, Osmanlı and S- enyuva, before and after model calibration. The three-dimensional finite element
24 November 2010
models of these two arch bridges, built in the ANSYS finite element program, are used to predict bridge
Accepted 10 December 2010
dynamic characteristics, such as natural frequencies and mode shapes. Following the analytical study,
Available online 18 February 2011
ambient vibration tests were conducted to experimentally obtain dynamic characteristics of these two
Keywords: bridges. During ambient vibration tests, accelerometers were placed at several points on the bridge to
Masonry arch bridges collect the vibration response due to natural and operational excitation sources. Enhanced frequency
Ambient vibration testing
domain decomposition and stochastic subspace identification techniques were used to extract the
Dynamic characteristics
experimental natural frequencies, mode shapes and damping ratios. Finite element models of the two
Earthquake behavior
Historical construction arch bridges were adjusted such that the model predictions reproduce the ambient vibration test results
Finite element modeling with increased fidelity. The behavior of the masonry arch bridges under earthquake excitation recorded
Model calibration during the Erzincan Earthquake in 1992 is simulated by both the initial and adjusted finite element
models. The findings of this study emphasize the importance of model calibration and ambient vibration
testing.
& 2010 Elsevier B.V. All rights reserved.

1. Introduction these operational conditions to ensure proper and effective con-

servation, restoration and maintenance, a reliable assessment of the
Masonry arch bridges, the oldest railway infrastructure elements structural behaviors of these bridges is critical.
in existence, still serve as a major component of this system in many It is of the utmost importance to ensure the safety of these
countries. For instance, Ellicks [1] noted that in the UK about 96% of historical bridges against the dynamic loadings of traffic, wind and
the 35,000 in-service masonry arch bridges at the time were over 70 earthquakes. This task requires the accurate identification of the
years old and 20% of these inventories were carrying principal roads. dynamic characteristics of these existing bridges (natural frequen-
Masonry arch bridges, which are in use worldwide, encompass cies, mode shapes and damping ratios). Finite element (FE) models
various sizes, styles and spans, some of which date back to early based on technical design data and engineering judgments were
antiquity. For example, the six-arch stone bridge 30 m in span built thought to provide reliable simulations of both the static and
around 100 BC is still in existence in Alcantra, Spain [1]. Such bridges dynamic behavior of arch bridges [5]. However, during the analy-
are an integral component of the heritage of cultures worldwide, and tical studies, unavoidable uncertainties occur when defining mate-
their preservation is of the utmost importance [2,3]. In Turkey, rial properties, boundary conditions and section areas of the design.
approximately 1300 of such historical bridges are still in-service. The These uncertainties limit the effectiveness of the FE method as a sole
first of these were built in the time of the Hittite, with subsequent determinant of the dynamic characteristics of an existing structure.
construction continuing during the Ottoman Empire [4]. These The FE method must be coupled with experimental methods while
bridges originally built for social and economical and strategic identifying natural frequencies, mode shapes and damping ratios of
purposes were intended for only pedestrian and horse-carriage historical bridges.
traffic. However, currently a significant number are being used for There are two experimental methods available to determine
motorized traffic. For the appraisal of their performance under new dynamic characteristics of a structure, the Experimental Modal
Analysis (EMA) and the Operational Modal Analysis (OMA). In EMA,
typically the acceleration response of the structure, caused by a
known excitation force generated by a shaker, hammer or other
n
Corresponding author at: Gümüs-hane University, Department of Civil
Engineering, 29000 Gümüs-hane, Turkey. controlled excitation devices, is measured. However, the need for
E-mail address: [email protected] (B. Sevim). such a controlled excitation source may render this method more

0168-874X/$ - see front matter & 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.finel.2010.12.011
622 B. Sevim et al. / Finite Elements in Analysis and Design 47 (2011) 621–634

expensive as compared to OMA. When applied to large-scale be more suitable for identifying dynamic characteristics of histor-
historic masonry monuments, this method is also inadequate for ical structures than EMA.
evenly and uniformly exciting the structure with a localized high FE models of the historical bridges are developed according to
amplitude excitation source. The required force levels to excite the existing engineering bridge drawings. Measurement tests,
global modes are typically high enough to induce nonlinear performed to validate the FE model predictions, typically yield a
response, which violates the fundamental linearity assumption level of disagreement from the expected results of the analytical
of EMA and tends to degrade the measurements. High excitation model. These discrepancies originate from the uncertainties and
sources can also cause localized damage on the historical con- simplifying assumptions in structural geometry and materials, as
struction. However, with the OMA method, the structure is vibrated well as from the inaccurate representation of boundary conditions.
by environmental and operational loads as traffic, wind or earth- The problem of adjusting the analytical model such that it
quake, with only the output responses being measured. OMA is also reproduces the experimental measurements with increased agree-
cheaper and faster because it obviates the need for excitation ment is known as model calibration in structural dynamics [6]. The
equipment. Moreover, since the vibration response under environ- sole purpose of model calibration is to improve the uncertain model
mental and operational loads is measured, OMA does not interfere parameters or imprecise modeling assumptions such that the FE
with the functional operations of the system. In addition, the model predictions are closer representations of reality.
measured response in OMA is representative of the real operating Many analytical and experimental studies have been conducted
conditions of the structure. Therefore, the authors believe OMA to on the historical bridges in the pertinent literature [7–10]. Beside
these studies, Fanning and Boothby [11] reported the results of field
testing and FE modeling of three masonry arch bridges. The bridges
were tested using a reference frame constructed beneath the
bridge structure, installing linear variable differential transformers
(LVDT) on the reference frame to measure structural displacements,
and loading the structure with a vehicle of known weight. Nonlinear
FE models were analyzed using ANSYS software. It was found that 3D
nonlinear finite element analysis using a reasonable set of material
properties enable good prediction of the actual behavior of a
masonry arch bridge. Frunzio et al. [12], in their three-dimensional
FE analysis of a stone masonry arch bridge involving nonlinear
material behavior, found that the results of their FE analysis were
useful in generating a qualitative map of the intervention areas for
restoration. They also found this analysis strongly dependent upon
the exactness of mechanical parameters, which are often difficult to
evaluate by experimental analyses, especially in cases of monu-
Fig. 1. Aerial views of the Osmanlı Arch Bridge. ments and historical buildings. Toker and Unay [13] studied the

Fig. 2. Geometrical properties of the Osmanlı Arch Bridge.

 B. Sevim et al. / Finite Elements in Analysis and Design 47 (2011) 621–634 623

mathematical modeling techniques on a prototype model of a The initial FE model predictions yield an on-average 10% disagree-
common arch bridge under different loading conditions. Ural [14] ment for Osmanlı Arch Bridge and on-average 15% disagreement for
determined the dynamic characteristics of Cos- andere Historical S- enyuva Arch Bridge. Because the deviations are believed to be the
Arch Bridge using SAP200 software. They also performed earthquake consequence of an imprecise modeling decisions or inaccurate
analysis of the bridge using obtained Elcentro ground motion record model parameters, the FE models of the bridges were calibrated
and maximum principal stresses. Bayraktar et al. [15] determined based on experimental natural frequencies. The effect of this
the dynamic characteristics of Historical Sinik Bridge under ambient calibration exercise of the FE model solutions was investigated
vibrations, and also updated the bridge FE model by adjusting the by comparing the earthquake behaviors of the bridges predicted by
boundary condition definitions. Brencich and Sabia’s [16] study of the initial and calibrated FE models.
Tanaro Bridge involved investigation of the in-service conditions and
the different stages of its demolition. In this study, the natural
frequencies, mode shapes and damping ratios of this 18 span
2. Formulations
masonry construction were identified by dynamic tests. Diamanti
et al. [17] used non-destructive ground-penetrating radar (GPR) on
In the absence of input force measurements, ambient excitation
masonry arch bridges for monitoring of ring separation. In order to
does not lend itself to frequency response functions (FRFs) or
validate and update the analytical results, several laboratory experi-
impulse response functions (IRFs) calculations. Therefore, a modal
ments were conducted. Analytical modeling techniques were used
identification procedure is based solely upon output data [18].
to simulate GPR tests, and the analytical models were updated using
Several modal parameter identification techniques are available for
laboratory experiments.
extracting these modal parameters, all of which were developed
This paper details how the earthquake behaviors of two histor-
due to improvements in computing capacity and signal processing
ical arch bridges (Osmanlı and S- enyuva Bridges) constructed in the
procedures. These approaches are the Operating Vectors Method,
Black Sea region of Turkey were determined using the authors’
the Complex Exponential Method, the Polyreference Time Domain
initial and calibrated FE models. The characteristics and geometric
Method, the enhanced frequency domain decomposition (EFDD)
properties of the case study bridges are first summarized, and the
and various stochastic subspace identification (SSI) techniques. In
development of the initial FE models is then described along with
this study, dynamic characteristics are extracted from experimen-
the initial predictions of bridge characteristics. Finally, the planning,
tal measurements using the last two in our list, the Enhanced
execution and outcome of ambient vibration tests of the two bridges
frequency domain decomposition (EFDD) and stochastic subspace
is discussed. From the measurements, the modal parameters are
identification (SSI) techniques.
identified by using both enhanced frequency domain decomposi-
tion (EFDD) andsubspace structural identification (SSI) techniques.
2.1. Enhanced frequency domain decomposition (EFDD) technique

In the EFDD technique, the relationship between the unknown

input x(t) and the measured responses y(t) is expressed as [19–21]

½Gyy ðjoÞ ¼ ½HðjoÞ ½Gxx ðjoÞ½HðjoÞT ð1Þ

where Gxx ðjwÞ is the power spectral density (PSD) matrix of the
input, Gyy ðjwÞ is the PSD matrix of the responses, HðjwÞ is the
frequency response function (FRF) matrix, and  and superscript T
denote complex conjugates and transpositions, respectively. The
Fig. 3. 3D finite element model of the Osmanlı Arch Bridge. solution to Eq. (1) is given in detail in the literature [22].

Fig. 4. Analytical mode shapes of the Osmanlı Arch Bridge.

624 B. Sevim et al. / Finite Elements in Analysis and Design 47 (2011) 621–634

2.2. Stochastic Subspace Identification (SSI) method 3. Finite element modeling and analytical dynamic
characteristics
Stochastic Subspace Identification (SSI) is an output-only
method that directly works with time data, circumventing the Both the Osmanlı and S- enyuva arch bridges are located in the
need to convert the time domain measurements to cross or auto- Fırtına Stream within the city limits of Rize, on the Black Sea Coast
correlations or to frequency spectra. The method is especially of north-eastern Turkey. The main structural elements of both
suitable for operational modal parameter identification, and the bridges are the stone arches, side walls and timber block. The
reader is referred to the following references for detailed technical bridges have not been subject to a restoration study.
overview [23–25]. Commercially available ANSYS v. 12 [26] software was used all
The model of vibrating structures can be defined by a set analytical modeling and bridge analyses. This software is prefer-
of linear, constant coefficient and second-order differential equa- able due to its capacity for linear, nonlinear, static and dynamic
tions [24]: analyses and its myriad of element types.

M U€ ðtÞ þ C U_ ðtÞ þKUðtÞ ¼ FðtÞ ¼ B uðtÞ ð2Þ

where M, C , K are the mass, damping and stiffness matrices, F(t) is 3.1. Osmanlı Arch Bridge
the excitation force, and U(t) is the displacement vector at
continuous time (t). Observe that the force vector F(t) is factorized Several views of the 19th century Osmanlı Arch Bridge are
into a matrix B describing the inputs in space and a vector u(t). In shown in Fig. 1.
the SSI method, the equation of dynamic equilibrium (Eq. (2)) is This two-spanned arch bridge has a total length of 51.7 m. The
converted to the more suitable discrete-time stochastic state-space span of each arch is 25.2 and 6 m, and the radius of each is 13 and
model [24]. Though this state-space model originates from control 3 m, respectively (see Fig. 2). The arches consist of inner and outer
theory, it is also useful in mechanical/civil engineering for comput- segments with thicknesses of 0.58 and 0.15 m, respectively. The
ing the modal parameters of a structure with a general viscous thickness of the side walls is 0.5 m with a timber block between
damping model [19]. them, the width of which is 2.50 m. Geometrical properties of the
bridge are provided in Fig. 2.
For the FE bridge analysis, three different material properties
were considered: the stone arch, the side walls and the timber
block. Identification of these material properties for historical
structures was completed with respect to the related successful
studies on other historical bridges [12,13,16]. The material prop-
erty values compiled from the literature and used in this analysis
are given below.
The solid model, the material types and the mesh of FE model of
the bridge is shown in Fig. 3. A total 38,602 SOLID186 elements are
Fig. 5. Aerial views of the S- enyuva Arch Bridge. used. Each SOLID186 element has 20 nodes and each node has three

Fig. 6. Geometrical properties of the S- enyuva Arch Bridge.

 B. Sevim et al. / Finite Elements in Analysis and Design 47 (2011) 621–634 625

degree of freedoms: x, y and z translations. The bridge abutments 3.2. -Senyuva Arch Bridge
and the side walls were constructed on the rock as shown in Fig. 1.
Therefore, all degrees of freedom under the bridge abutments and Fig. 5 shows the second subject bridge of our study, the 17th
at the side walls are assumed as fixed. century S- enyuva Arch Bridge.
The first five natural frequencies have been estimated to range Built in 1696, the single-arch S- enyuva Bridge spans the Fırtına
between 3 and 15 Hz. The bridge has three main types of vibration (Storm) River at a height of 12.4 m and a length of 24.8 m. The total
modes: bending modes in z direction, vertical modes in y direction length of the bridge is 52.4 m and the width of the deck is 2.5 m. The
and torsional modes. FE model predictions for the first five mode geometrical properties of the S- enyuva Bridge are shown in Fig. 6.
shapes of the bridge and their corresponding natural frequency The radius and thickness of this magnificent historical structure is
predictions are shown in Fig. 4. 12.4 and 0.60 m, respectively. The thickness of the side walls is
0.50 m and there is a timber block between them, which is 1.5 m in
width. The height of both side walls are 9.2 and 3.5 m, respectively.
The bridge also has 0.30  0.60 m2 dimensional parapets on both
the sides of the bridge deck.
Fig. 7 shows a 3D FE model of the bridge, built with a total 26,408
SOLID186 elements. The materials properties used in the analyses
are listed in Table 1. The boundary conditions are defined by fixing
the translational and rotational degrees of freedom at all bridge
abutments and both side walls.
The first five natural frequencies are predicted to fall within a
range of 3–11 Hz. The first five mode shapes of the bridge, shown in
Fig. 8, are bending modes in the z direction, vertical modes in the y
direction and torsional modes.

4. Operational modal analysis and identification of

experimental dynamic characteristics

Uni-axial accelerometers (B&K 8340) along with a 17-channel

data acquisition system (B&K 3560) were used during the ambient
measurements of the bridges. Ambient vibration tests were con-
Fig. 7. Finite element model of the S- enyuva Arch Bridge.
ducted under existing environmental and operational loads as

Table 1
Material properties considered in the analytical solution.

Material Modulus of Poisson ratio Density

elasticity (N/m2) (kg/m3)

Stone arches 3.0E9 0.25 1600

Timber block 1.5E9 0.05 1300
Side walls 2.5E9 0.20 1400
Fig. 9. Equipment used during ambient vibration testing.

Fig. 8. Analytical mode shapes of the S- enyuva Arch Bridge.

626 B. Sevim et al. / Finite Elements in Analysis and Design 47 (2011) 621–634

wind and human movement. To generate the detectable vibration testing difficulties (e.g. inadequate accelerometers, very rigid
levels for our study, between five and ten people were directed to structure, inadequacy vibration signals), the first five modes were
walk across the bridge. This procedure, known as human move- only considered in our experimental measurements.
ment and is used in the literature. The minimum frequency span
and sensitivity of these accelerometers reign were assigned as
0.1 Hz and 10 V/g, respectively, and signals were transferred into 4.1. Osmanlı Arch Bridge
the PULSE [27] Lapshop software. The data acquisition system
and accelerometers are pictured in Fig. 9. The operational modal Fifteen uni-axial accelerometers were used in the ambient
analysis software (OMA) [28] was used to generate parameter vibration tests of the bridge. The 3D schematic view of the
estimations from the ambient vibration data. Because of various accelerometer placement is given in Fig. 10.

Fig. 10. Accelerometer locations on the Osmanlı Arch Bridge.

Fig. 11. Modal parameters of Osmanlı Bridge attained from EFDD and SSI techniques: (a) singular values of spectral density matrices and (b) stabilization diagram of estimated
state space models.
 B. Sevim et al. / Finite Elements in Analysis and Design 47 (2011) 621–634 627

The OMA was conducted using the EFDD technique in the 5. Finite element model calibration of the bridges
frequency domain and the SSI technique in the time domain. The
EFDD technique was used to extract dynamic characteristics from The natural frequencies obtained by the initial FE model of the
singular values of each vibration signal, while SSI technique was two bridges show differences from the experimental results: an
used to compute the singular values of a collection of vibration approximate 10% difference for the Osmanlı Arch Bridge (Table 2),
signals. Fig. 11 shows the singular values of the spectral density and an approximate 15% difference for S- enyuva Arch Bridge
matrices and the stabilization diagrams of estimated state space (Table 3). Because we hypothesized that these deviations were
models obtained from vibration signals using these two techni- the result of imprecision in the FE model, the FE models of these
ques. This step consists of computation of singular values. Repeat- bridges were calibrated according to the experimental results.
ing the procedure with increasing polynomial orders yields a The main purpose of the whole paper is to illustrate the
stabilization diagram, which plots the roots of the polynomial as importance of model calibration. Therefore, the material property
the order of the polynomial is increased. The stable roots for values used in the FE model are determined according to pertinent
increasing polynomial orders indicate a natural system frequency, literature for masonry structures obtained in the laboratory and
and the user defines the maximum number of modes for the in-situ tests. However, the actual material properties of the bridges
analysis, which corresponds to the maximum polynomial order. must be obtained from laboratory and in-situ tests when consider-
Typically, an over-specified mode order is defined, which later ing earthquake performance for strengthening. Despite unavoid-
necessitates a judgment-based distinction between the physical able uncertainties in the material properties, the material property
and computational modes. The stabilization diagram aids this definitions are considered to higher confidence. Therefore, the
process. The premise is that as these physical (actual) poles manual calibration exercise, described in the next section, will
stabilize for an increasing model order, the mathematical poles primarily focus on the boundary condition definitions. The goal of
scatter. Though this pole stabilization is apparent in Fig. 11, the pole model calibration as implemented herein is to ‘improve’ the FE
scattering is less clearly defined. The reason for this apparent model, not to obtain a FE model that identically matches the
discrepancy is the lack of appropriate signals collected for mode experimental results.
extraction.
The first five mode shapes obtained from ambient vibration
tests are shown in Fig. 12. A visual comparison of Figs. 8 and 12
5.1. Osmanlı Arch Bridge
show good agreement between analytical and experimental mode
shapes. A comparison of analytically and experimentally identified
As shown in Fig. 16, some river materials such as soil, sand, rock and
natural frequencies and damping ratios (x) of Osmanlı Arch Bridge
rubbish are collected at the side walls of Osmanlı Arch Bridge, which
is provided in Table 2.
have constrained the displacement of the side walls over time.
Therefore, the sidewall stiffness was increased in the calibrated model.
4.2. -Senyuva Arch Bridge
Table 2
Comparison of analytically and experimentally identified dynamic characteristics of
Twelve uni-axial accelerometers were used in the ambient the Osmanlı Arch Bridge.
vibration tests of the bridge. The 3D schematic view of location
points of the accelerometers is shown in Fig. 13. The locations of the Mode Analytical Experimental Damping
accelerometers are determined according to finite element results. number frequencies frequencies (Hz) (%)
(Hz)
Singular values of spectral density matrices and stabilization
EFDD SSI f
diagrams of estimated state space models attained from vibration
signals using EFDD and SSI techniques are shown in Fig. 14. 1 3.843 4.640 4.642 1.634
The first five mode shapes extracted from ambient vibration 2 7.527 8.094 8.325 1.035
tests are given in Fig. 15. A comparison of analytically and 3 9.371 9.879 9.735 6.157
4 10.638 12.340 11.910 0.256
experimentally identified dynamic characteristics, and the damping 5 14.563 15.840 15.420 0.159
ratios (x) of the S- enyuva Bridge is given in Table 3.

Fig. 12. Experimental mode shapes of the Osmanlı Arch Bridge.

628 B. Sevim et al. / Finite Elements in Analysis and Design 47 (2011) 621–634

Table 4 shows a comparison of experimental and analytical 5.2. -Senyuva Arch Bridge
natural frequencies after adjusting the boundary conditions in the
FE model. It can be seen that the differences between natural Similar to the Osmanli bridge, the displacement ability of the
frequencies are reduced from on average 15% to about on average walls near the both sides of the S- enyuva Bridge are decreased due to
5% after model calibration. Considering the two different techni- piling up the earth, stone, rubbish (Fig. 17). As a result, the stiffness
ques used in the identification of modal parameters, EFDD and SSI, of the side walls was increased in the calibrated model.
yielded up to 3% difference, on average 5% variability between Table 5 shows a comparison of the experimental and analytical
measured and calculated natural frequencies is deemed acceptable. natural frequencies after FE model calibration. Note that the
differences between natural frequencies are approximately 2%
after model calibration, which is in the same order of magnitude
of difference between the frequencies identified by EFDD and SSI
methods.

6. Earthquake behaviors of Arch Bridges

The actual seismic performance of existing masonry arch

bridges must be evaluated by using nonlinear materials and
considering the interface between the masonry and the fill. Because
the purpose is to illustrate the importance of model calibration and
ambient vibration testing on the earthquake performance of
masonry arch bridges, linear earthquake behaviors of Osmanlı
and S- enyuva Arch Bridges were considered to investigate the effect
of FE model calibration. However, the study presented herein can
be followed up with a nonlinear study.
Time-history analyses of the two bridges of interest were
performed using ERZIKAN/ERZ-NS component of 1992 Erzincan
Fig. 13. Accelerometer locations on the S- enyuva Arch Bridge. earthquake (Fig. 18) [29]. The earthquake occurred in the North

Fig. 14. Modal parameters of the S- enyuva Bridge from EFDD and SSI techniques.
 B. Sevim et al. / Finite Elements in Analysis and Design 47 (2011) 621–634 629

Fig. 15. Experimental mode shapes of the S- enyuva Arch Bridge.

Table 3 Table 4
Comparison of analytically and experimentally identified dynamic characteristics of Analytical and experimental natural frequencies of the Osmanlı Arch Bridge after
the S- enyuva Arch Bridge. model calibration.

Mode Analytical Experimental Damping Mode Analytical Calibrated Experimental

number frequencies frequencies (Hz) (%) number frequencies analytical frequencies (Hz)
(Hz) (Hz) frequencies
EFDD SSI f (Hz) EFDD SSI

1 3.347 4.045 4.066 2.377 1 3.843 4.640 4.640 4.642

2 5.772 7.750 7.960 1.318 2 7.527 8.757 8.094 8.325
3 7.554 8.020 8.044 4.288 3 9.371 9.875 9.879 9.735
4 9.055 10.000 10.100 0.265 4 10.638 11.967 12.340 11.910
5 10.044 12.160 11.750 0.345 5 14.563 15.703 15.840 15.420

Fig. 16. Side views of the side walls of the Osmanlı Arch Bridge believed to restrict
wall displacement ability.
Fig. 17. View of the S- enyuva Arch Bridge side walls.

Anatolian Fault, which is the nearest fault to the bridges that were 6.1. Osmanlı Arch Bridge
the subject of this work. The accelerations are applied to bridges on
bending direction (in z direction) where the first mode shapes are The maximum displacement contours of the Osmanlı Arch Bridge
obtained. before and after model calibration are shown in Fig. 19(a) and (b),
In the time-history analyses, element mass and stiffness respectively. These contours represent the distribution of the peak
matrices are computed using the Gauss numerical integration values reached by the maximum displacements at any time during
technique [30]. The Newmark method is used in the solution of the simulation at each point within the section. Note that the maximum
equation. Because of the computational demand of this method, displacements occur at the middle region of the bridge, top of the
only the first 6.5 s of the earthquake were used during calculations. timber block and side walls (Fig. 18). Experimental damping ratios
Such an abbreviated duration of time was not expected to adversely are used and the stiffness of the side walls is increased in the
affect results as the first few seconds of the Erzincan Earthquake updated model.
were the most effective (Fig. 18). The time histories of displacements at the nodal point of stone
Because the damping ratios are unknown in the initial (uncali- arch in which the maximum displacements occurred are plotted
brated) FE model analysis, the authors estimated the Rayleigh before and after model calibration in Fig. 20(a) and (b), respectively.
damping coefficients for an assumed 5% damping ratio. However, The maximum displacements, obtained from the initial FE model
during the calibrated FE model analysis, Rayleigh damping coeffi- (prior to model calibration), were at an acceptable level [31].
cients were calculated according to the experimentally obtained The maximum principal stress contours of Osmanlı Arch Bridge
damping ratios. before and after model calibration are shown in Fig. 21(a) and (b),
630 B. Sevim et al. / Finite Elements in Analysis and Design 47 (2011) 621–634

respectively. These stress contours represent the distribution of the The time histories of the maximum and minimum principal stresses
peak values reached by the maximum principal stress at each point are plotted before and after model calibration in Fig. 22(a)–(d),
within the section. It can be seen in Fig. 21 that the maximum respectively. These stresses occur at the heel point of the bridge
principal stresses are occurred displacements at any time during (Fig. 21). As seen in Fig. 22, the principal stresses obtained from initial
simulation at the region around the base of stone arch. The model (before model calibration) are significantly larger than those of
maximum and minimum principal stress values in the arch are the calibrated model.
lower than the acceptable strength of the stone [11,32].

6.2. -Senyuva Arch Bridge

Table 5
Analytical and experimental natural frequencies of the S- enyuva Arch Bridge after The maximum displacement contours of the S- enyuva Arch Bridge
model calibration. before and after model calibration are shown in Fig. 23(a) and (b),
respectively. These contours represent the distribution of the peak
Mode Analytical Calibrated Experimental
values reached by the maximum displacements at each point within
number frequencies analytical frequencies (Hz)
(Hz) frequencies the section. It can be seen from Fig. 23 that the maximum displace-
(Hz) EFDD SSI ments occurred at the middle region of the bridge and the top of the

1 3.347 4.070 4.045 4.066

2 5.772 7.780 7.750 7.960
3 7.554 8.020 8.020 8.044 Before model calibration After model calibration
4 9.055 10.670 10.000 10.100
5 10.044 12.090 12.160 11.750
30 30
20 20

Displacements (mm)

Displacements (mm)
10 10
Acceleration (m/s2)

6.0
Max=0.515g
3.0 0 0

0.0 -10 -10

-3.0 -20 -20
-6.0 -30 -30
0 1 2 3 4 5 6 7 8 9 10 11 12 13 0 1 2 3 4 5 6 0 1 2 3 4 5 6
Time (s) Time (s) Time (s)

Fig. 18. The time-history of ground motion acceleration of the 1992 Erzincan Fig. 20. The time histories of displacements at stone arch of the Osmanlı Arch
Earthquake. Bridge.

Fig. 19. The maximum displacements contours of the Osmanlı Arch Bridge obtained before/after model calibration.
 B. Sevim et al. / Finite Elements in Analysis and Design 47 (2011) 621–634 631

Fig. 21. The maximum principal stress contours of the Osmanlı Arch Bridge before and after model calibration.

timber block and side walls. It can be generally stated that the was investigated using both the initial (not calibrated) and
displacements are at an acceptable level [31]. calibrated FE models. The following phenomena were observed:
The time histories of displacements at the nodal point of stone
arch where the maximum displacements occurred are plotted
before and after model calibration in Fig. 24(a) and (b), respectively.  Initial analytical natural frequencies of Osmanlı Arch Bridge were
The maximum vales in time histories of displacements are obtained attained at ranges between 3.8 and 14.6 Hz for the first five modes.
for initial FE model (before the model calibration). These can be classified into bending modes in the z direction,
The maximum principal stress contours of S- enyuva Arch Bridge vertical modes in the y direction and torsional modes. The first five
before and after model calibration are shown in Fig. 25(a) and (b), experimental modes were estimated within ranges between
respectively. These stress contours represent the distribution of the 4.6 and 15.8 Hz. The experimental mode shapes were observed
peak values reached by the maximum principal stress at each point to be in close agreement with the analytical mode shapes.
within the section. Similar to the Osmanli bridge, maximum  For Osmanlı Arch Bridge, there is an approximate 10% difference
principal stresses are observed at the base of stone arch before between the natural frequencies predicted by the initial FE
the model calibration (Fig. 21). However, after calibration, the model and obtained through OMA. The FE model of Osmanlı
predicted maximum stress region shifted to the left upside of the Arch Bridge is calibrated by adjusting the FE model boundary
side wall. The maximum and minimum principal stress values in conditions. After model calibration, the differences in calculated
the arch are lower than the stone strength [11,32]. and measured natural frequencies were reduced to about 5%.
The time histories of the maximum and minimum principal stresses  The obtained initial analytical natural frequencies of S- enyuva
are plotted before and after model calibration in Fig. 26(a)–(d), Arch Bridge ranged between 3.4 and 10 Hz for the first five
respectively. These stresses occur at the heel point of the bridge modes. These modes can be classified into bending, vertical and
(Fig. 25). As seen in Fig. 26, the principal stresses obtained from initial torsional modes. The first five experimental modes were within
model (before model calibration) are significantly larger than those of a range between 4.1 and 12.1 Hz. Experimental mode shapes
the calibrated model. were visually comparable to analytical mode shapes.
 There is an approximate 15% difference between the initial FE
model and the experimental natural frequencies of the S- enyuva
7. Conclusions Arch Bridge. The FE model of the S- enyuva Arch Bridge was also
calibrated by modifying boundary conditions around the side
In this paper, FE modeling, modal testing, FE model calibration walls where there was an accumulation of debris. After model
of Osmanlı and S- enyuva Historical Arch Bridges located in Rize, calibration, the discrepancy between the FE model and mea-
Turkey are presented. The earthquake behavior of the two bridges surements was observed as low as 2%.
632 B. Sevim et al. / Finite Elements in Analysis and Design 47 (2011) 621–634

 The maximum displacements, obtained from the earthquake displacements from the earthquake analyses of initial and
analyses of initial and calibrated models of Osmanlı Arch calibrated models of S- enyuva Arch Bridge were as 0.039 and
Bridge, were 0.022 and 0.016 m, respectively. The maximum 0.025 m, respectively. The maximum displacements occurred at
the middle region of the bridges and at the top of the timber
block and side walls.
Max. stresses before Max. stresses after  It can be generally stated that the displacements are at an
model calibration model calibration acceptable level.
Maximum Principal Stress (MPa)

Maximum Principal Stress (MPa)

3 3  The maximum principal stresses of 2.6 and 1.2 MPa were
obtained, respectively, from earthquake analyses of initial and
calibrated models of the Osmanlı Arch Bridge. The maximum
2 2 principal stresses of 2.7 and 1.6 MPa were obtained, respectively,
from earthquake analyses of initial and calibrated models of the
S- enyuva Arch Bridge. These maximum principal stresses gen-
1 1 erally occurred at regions around the base of stone arch for both
bridges.
 The minimum principal stresses of 2.4 and 1.09 MPa were
0 0 obtained, respectively, from earthquake analyses of the initial
0 1 2 3 4 5 6 0 1 2 3 4 5 6 and calibrated models of the Osmanlı Arch Bridge. The mini-
Time (s) Time (s) mum principal stresses of 2.7 and 1.6 MPa were obtained,
Min. stresses before Min. stresses after
model calibration model calibration
Minimum Principal Stress (MPa)

Minimum Principal Stress (MPa)

Before model calibration After model calibration

0 0
40 40
Displacements (mm) 30 30

Displacements (mm)
-1 -1 20 20
10 10
0 0
-2 -2 -10 -10
-20 -20
-30 -30
-3 -3
-40 -40
0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6
Time (s) Time (s) Time (s) Time (s)

Fig. 22. Time histories of (a), (b) maximum and (c), (d) minimum principal stresses Fig. 24. The time histories of the stone arch displacements of the S- enyuva Arch
of the Osmanlı Arch Bridge. Bridge.

Fig. 23. The maximum displacement contours of the S- enyuva Arch Bridge obtained before/after model calibration.
 B. Sevim et al. / Finite Elements in Analysis and Design 47 (2011) 621–634 633

Fig. 25. The maximum principal stress contours of the S- enyuva Arch Bridge before and after model calibration.

Max. stresses before model Max. stresses after model  The maximum and minimum results obtained from the initial
calibration calibration models are bigger than those of the calibrated models for both
the Osmanlı and S- enyuva Arch Bridges.
Maximum Principal Stress (MPa)

Maximum Principal Stress (MPa)

3 3
 These findings illustrate a severe overestimation of the uncali-
brated FE model constructed and informed with the best-
2 2 engineering practices. Therefore, an ambient testing and model
calibration scheme must be an integral part of routine structural
analyses when applied to historic masonry arch bridges.
1 1

Acknowledgements
0 0
This research was supported by the TUBITAK and Karadeniz
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Technical University under Research Grant nos. 106M038,
Time (s) Time (s)
2005.112.001.1, and 2006.112.001.1, respectively. Editorial help
Min. stresses before model Min. stresses after model of Godfrey Kimball of Clemson University is greatly appreciated.
calibration calibration
Minimum Principal Stress (MPa)

Minimum Principal Stress (MPa)

0 0 References

[1] C. Ellicks, Ph.D. Thesis, The University of Reading, the U.K., 1979.
-1 -1 [2] I.H.P. Mamaghani, Analysis of masonry bridges: discrete Finite Element
Method, Transportation Research Record 76 (2006) 13–19.
[3] R.O. Catalan, L.E. Aldea, New tendencies on repair and strengthening on
masonry arch bridges, in: 5th International Conference on Arch Bridges, 12–14
-2 -2 September, Madeira, Portugal, 2007, pp. 719–724.
[4] Url-1, http://www.kgm.gov.tr/asps/KGM/koprucalisma/tarihikopruler.pdf, (2009).
[5] W.X. Ren, X. Penga, Y. Lina, Experimental and analytical studies on dynamic
characteristics of a large span cable-stayed bridge, Engineering Structures 27
-3 -3 (4) (2005) 535–548.
0 1 2 3 4 5 6 0 1 2 3 4 5 6 [6] M.I. Friswell, J.E. Mottershead, Finite Element Model Updating in Structural
Time (s) Time (s) Dynamics, Kluwer Academic Publishers, Dordrecht, The Netherlands, 1995.
[7] J. Page, Masonry Arch Bridges—A State of the Art Review, HMSO, London, 1993.
Fig. 26. Time histories of (a) and (b) maximum and (c) and (d) minimum principal [8] D.M. Armstrong, A. Sibbald, C.A. Fairfield, M.C. Forde, Modal analysis for
masonry arch bridge spandrel wall separation identification, NDT & E Inter-
stresses of the S- enyuva Arch Bridge.
national 28 (6) (1995) 377–386.
[9] A. Bensalem, C.A. Fairfield, A. Sibbald, Non-destructive evaluation of the
dynamic response of a brickwork arch, ICE Journal of Structures and Buildings
respectively, from earthquake analyses of initial and calibrated 122 (1) (1997) 69–82.
models of the S- enyuva Arch Bridge. [10] A. Bensalem, C.A. Fairfield, A. Sibbald, Damping effects on the NDT of soil
backfilled arch bridges, Journal of British Institute NDT 40 (2) (1998) 107–116.
 The maximum and minimum principal stress values in the arch [11] P.J. Fanning, T.E. Boothby, Three-dimensional modelling and full-scale testing
are lower than the stone strength. of stone arch bridges, Computers and Structures 79 (29) (2001) 2645–2662.
634 B. Sevim et al. / Finite Elements in Analysis and Design 47 (2011) 621–634

[12] G. Frunzio, M. Monaco, A. Gesualdo, 3D FEM analysis of a roman arch bridge, modal analysis, in: Proceedings of the ISMA2006: International Conference on
Historical Constructions (2001) 591–598. Noise and Vibration Engineering, Leuven, Belgium, 2006.
[13] S. Toker, A.I. Ünay, Mathematical modelling and finite element analysis [22] R. Brincker, L. Zhang, P. Andersen, Modal identification from ambient responses
of masonry arch bridges, Journal of Science of Gazi University 17 (2) (2004) using frequency domain decomposition, 18th International Modal Analysis
129–139. Conference, San Antonio, USA, vol. 4062(2), 2000, pp. 625–630.
[14] A. Ural, Finite element analysis of historical arch bridge, in: International Earth- [23] P. Van Overschee, B. De Moor, Subspace Identification for Linear Systems:
quake Symposium Kocaeli, Kocaeli, Turkey, 23–25 March 2005, pp. 408–413. Theory–implementation–applications, Kluwer Academic Publishers, Dordrecht,
[15] A. Bayraktar, A.C. Altunıs-ık, T. Türker, B. Sevim, The model updating of NL, 1996.
historical masonry bridges using operational modal analysis method, in: [24] B. Peeters, G. De Roeck, Reference based stochastic subspace identification in
Proceedings of the 1st National Conference Reinforcement and Transfer into civil engineering, in: Proceedings of the 2nd International Conference on
the Future of Historical Structures, Ankara, Turkey, 2007, pp. 429–440.
Identification in Engineering Systems, Swansea, UK, 1999, pp. 639–648.
[16] A. Brencich, D. Sabia, Experimental identification of a multi-span masonry
[25] B. Peeters, System identification and damage detection in civil engineering,
bridge: the Tanaro Bridge, Construction and Building Materials 22 (10) (2008)
Ph.D. Thesis, K.U., Leuven, Belgium, 2000.
2087–2099.
[26] ANSYS, Swanson Analysis System, USA, 2008.
[17] N. Diamanti, A. Giannopoulos, M.C. Forde, Numerical modelling and experi-
[27] PULSE, Analyzers and Solutions, Release 11.2, Bruel and Kjaer, Sound and
mental verification of GPR to investigate ring separation in brick masonry arch
Vibration Measurement A/S, Denmark, 2006.
bridges, NDT and E International 41 (5) (2008) 354–363.
[28] OMA, Release 4.0, Structural Vibration Solution A/S, Denmark, 2006.
[18] W.X. Ren, T. Zhao, I.E. Harik, Experimental and analytical modal analysis
[29] Url-2, Pacific Earthquake Engineering Research Center (PEER), /http://peer.
of steel arch bridge, Journal of Structural Engineering, ASCE 130 (7) (2004)
1022–1031. berkeley.edu/smcat/dataS, 2009.
[19] D.J. Ewins, Modal Testing: Theory and Practice, Research Studies Press Ltd., [30] K.J. Bathe, Finite Element Procedures in Engineering Analysis, Prentice-Hall,
England, 1984. Englewood Cliffs, New Jersey, 1996.
[20] J.S. Bendat, A.G. Piersol, Random Data: Analysis and Measurement Procedures, [31] L. Pela , A. Aprile, A. Benedetti, Seismic assessment of masonry arch bridges,
John Wiley and Sons, USA, 2004. Engineering Structures 3 (2009) 1777–1788.
[21] N.J. Jacobsen, P. Andersen, R. Brincker, Using enhanced frequency domain [32] K. Venu Madhava Rao, B.V. Venkatarama Redyy, K.S. Jagadish, Strength
decomposition as a robust technique to harmonic excitation in operational characteristic of stone masonry, Materials and Structures 30 (1997) 233–237.